Meta::MAgent Class Reference
Symbolic manipulation, function reduction, machine learning, code evolution, and stochastic functions. More...
#include <MAgent.h>
Detailed Description
Symbolic manipulation, function reduction, machine learning, code evolution, and stochastic functions.
MAgent is the preferred class to use for M interpretation. It provides symbolic manipulation, mathematical function reduction, and code evolution using genetic programming, neural networks, and other machine learning techniques. MAgent provides a limited knowledge base that may be utilized via Tell() and Ask().
In addition, the agent includes functions for interfacing with Mathematica if it is available on your machine: just set the "mMMPath" config to the path of your "math" command. Although the agent may be used to interface with Mathematica, no part of the MAgent implemenation makes use of or requires Mathematica -- except for those functions named MMxxx.
Functional versions of the stochastic methods in MRandom are made available here. MAgent is fully storeable and restoreable. In most cases, agent should be run in non-strict mode, see MObject::setStrict(). Exact mode may also be desired, see MObject::setExact().
Constructor & Destructor Documentation
| Meta::MAgent::MAgent | ( | ) |
Constructor.
| Meta::MAgent::MAgent | ( | const MAgent & | agent | ) |
Copy constructor.
| Meta::MAgent::MAgent | ( | const mvar & | v, | |
| RestoreOpt | opt | |||
| ) |
Restore constructor.
| Meta::MAgent::MAgent | ( | MBrokerBase * | broker | ) |
Remote object constructor - do not call directly.
| virtual Meta::MAgent::~MAgent | ( | ) | [virtual] |
Destructor.
Member Function Documentation
Query the agent.
- Parameters:
-
query e.g: LT(x,10)
- Returns:
- True or False (False if undetermined).
See MRandom::bernoulli().
See MRandom::binomial().
See MRandom::chisquare().
| virtual MObject* Meta::MAgent::clone | ( | ) | const [virtual] |
Return a copy of this agent.
Reimplemented from Meta::MObject.
| bool Meta::MAgent::configure | ( | const mvar & | args | ) |
Configure the agent based on args.
- Returns:
- true if successful.
Delete the specified objective function.
- Parameters:
-
f objective function, e.g: f(x,y)
See MRandom::erlang().
Explicitly evaluate an objective function. Usually this function does not need to be called directly.
Evolve an objective function. Upon completion the best solution is set to the objective function in the agent's scope.
- Parameters:
-
f The objective function, e.g: f(x,y) maxTime Maximum number of seconds to evolve for. targetError Stop upon reaching this error.
- Returns:
- The current error of the objective.
See MRandom::geometric().
See MRandom::lognormal().
Convert f to Mathematica function syntax without translating functions.
See MRandom::normal().
Neural-network based evolver using supervised learning.
- Parameters:
-
f The objective function to be evolved, e.g: f(x,y) train the training set, e.g: [[f:4.23954, x:0.97627, y:1.85689],[f:14.3957, x:4.30379, y:6.88531]] layers The number of hidden layers in the network. height The height of each hidden layer.
| mnode Meta::MAgent::Objective_ | ( | mnode | f, | |
| mnode | train, | |||
| mnode | prods, | |||
| mnode | population, | |||
| mnode | maxDepth, | |||
| mnode | evalTimeout = 5.0, |
|||
| mnode | mutationRate = 0.05, |
|||
| mnode | constMutationRate = 0.0, |
|||
| mnode | selectionPressure = 0.5 | |||
| ) |
Specify an objective to be evolved using genetic programming techniques. Supervised or unsupervised learning may be used.
- Parameters:
-
f The output function name to be evolved, e.g: f(x,y) train The training set of inputs and outputs, e.g: for supervised learning: [[f:4.23954, x:0.97627, y:1.85689],[f:14.3957, x:4.30379, y:6.88531]], for unsupervised learning: [[f:evalFunc(), x:0.97627, y:1.85689],[f:evalFunc(), x:4.30379, y:6.88531]]. prods The production rules / grammar, e.g: [[A,[A+A, A-A, A*A, A/A, B]],[B,[1,2,x,y]]], where production symbols are in all-caps and the first rule gives the start production. population The population size -- larger populations run slower and require more memory but provide greater diversity and potentially better solutions. If an objective fails to find an adequate solution, increasing population should be one of the first things to try. maxDepth Limits the complexity of the output solution. It determines the number of productions that can be selected before a terminal is chosen. evalTimeout The maximum number of seconds to evaluate each solution over all of its inputs for. If the evaluation times out, its error is set to infinity. mutationRate The mutation rate per population. 0..1.0 constMutationRate Constant mutation rate per population. Mutates floating point constants. 0..1.0 selectionPressure Higher values favor more strongly the selection of solutions with less error. 0..1.0
- Parameters:
-
f Objective function, e.g: f(x,y)
- Returns:
- The last error associated with evolving an objective function.
| mnode Meta::MAgent::Objectives_ | ( | ) |
- Returns:
- A listing of the objectives and their parameters.
See MRandom::pareto().
See MRandom::pascal().
Enumerate permutations, choosing k.
- Parameters:
-
f e.g: [1,2,3] k e.g: 2
- Returns:
- e.g: ([1,2],[2,1],[3,1],[1,3],[2,3],[3,2])
Enumerate permutations.
- Parameters:
-
f e.g: [1,2,3]
- Returns:
- e.g: ([1,2,3],[2,1,3],[3,1,2],[1,3,2],[2,3,1],[3,2,1])
See MRandom::poisson().
Process method.
Reimplemented from Meta::MObject.
Reduce a mathematical function.
- Parameters:
-
f e.g: 1 + y + 2*y + 3*z / 22*z + 99
- Returns:
- e.g: 100 + 3 * y + 3/22 * z^2
Identical to Sample(3) but make the output function g.
- Parameters:
-
g e.g: g(x,y)
Sample a function over the specified inputs ranges, for count times. Useful for generating an objective test training set.
- Parameters:
-
f e.g: f(x,y) ranges e.g: [[-10,10], [5,10]] count Number of samples to return.
- Returns:
- e.g: [[f:1.90472, x:0.97627, y:0.928446],[f:7.74644, x:4.30379, y:3.44266]]
Seed the agent's random number generator.
- Parameters:
-
seed is 0..18446744073709551615
| virtual void Meta::MAgent::store | ( | mvar & | v | ) | const [virtual] |
Store this agent to v.
Reimplemented from Meta::MObject.
Tell the agent a relation has the specified value whenever condition is true thereby executing the post function.
- Parameters:
-
condition Eq(x,2) relation e.g: Mul(x,y) value e.g: 8 post e.g: Inc(x)
- Returns:
- true on success.
Tell the agent a relation has the specified value whenever condition is true.
- Parameters:
-
condition e.g: Eq(x,2) relation e.g: Mul(x,y) value e.g: 8
- Returns:
- true on success.
Tell the agent that a relation has the specified value.
- Parameters:
-
relation e.g: Add(x,y) value e.g: 8
- Returns:
- true on success.
Tell the agent a fact.
- Parameters:
-
fact e.g: IsCommutative(Add)
- Returns:
- true on success.
| mnode Meta::MAgent::Time_ | ( | ) |
Return time since the UNIX epoch in seconds + microseconds.
| mnode Meta::MAgent::TimeSeed_ | ( | ) |
Seed the agent's random number generator with the current time.
See MRandom::uniform().
| mnode Meta::MAgent::Uniform_ | ( | ) |
See MRandom::uniform().
The documentation for this class was generated from the following file:
- AndroMeta/include/Meta/MAgent.h
